Implicit Bias in Letters of Recommendation for Plastic Surgery Residency Applicants
Alisha R. Bonaroti, MD, MS1; Wendy Chen, MD, MS2; Eva Bacas3; Na-Rae Han, PhD3; Scott F. Kiesling, PhD3; Lesley Wong, MD1; Debra A. Bourne, MD1
1Division of Plastic and Reconstructive Surgery, University of Kentucky, Lexington KY 2Department of Orthopedic Surgery, University of California Los Angeles, Los Angeles CA 3Department of Linguistics, University of Pittsburgh, Pittsburgh PA
Background: Implicit bias refers to a set of unconscious beliefs which affect one’s actions and decisions. The literature is fraught with examples of objective biases within the surgical discipline, however implicit biases can be more difficult to ascertain. Applications for surgical residency are primarily based on objective data, but are heavily weighted on subjective assessment through letters of recommendation. We aim to investigate the content of letters of recommendation for plastic surgery residency applicants to determine if any gender disparities exist.
Methods: All letters of recommendation submitted to a single integrated plastic surgery residency program during the 2019 – 2020 academic year were included. Linguistic comparisons were performed to assess for differences in applicant characterization. Lexical analysis was used to identify specific words which associated heavily with one gender.
Results: A composite of 658 letters (337 male, 51.2%; 321 female, 48.7%) written on behalf of 175 applicants (91 male, 52.0%; 84 female, 48.0%) were analyzed. A total of 57 female (12.3%) and 406 male (87.7%) unique letter writers were identified. Specific words associated with female applicants in order of frequency included “policy, media, judgement, oversight, weak, thoughtfully, domestic, resilience, precision, tough.” Words associated with male applicants in order of frequency included “reliability, gentleman, improvement, diversity, hardest, admirable, arrogant, dependability, flexible, happy, facilitator, integrates, stronger, supported, oversaw, demanding.” A machine learning model was developed with the ability to predict an applicant’s gender after removing all gender pronouns with 60.7% precision.
Conclusion: The gender profile of applicants was evenly divided, however a significantly higher proportion of letter writers were male. The association between specific objective descriptors and genders does imply an element of bias. This is further reinforced by a predictive model with precision which is higher than chance alone. Further study is indicated for computational text analysis accompanied by qualitative analysis of discourse. The continued study of implicit bias is fundamental to the field of plastic surgery as we continue to overcome both objective and subjective inequalities.
Back to 2021 Abstracts